Interesting take away by @yoavgo about #dlearn models for QA.pic.twitter.com/lYL4WDwNvN
-
Even memory networks by FAIR / neural reasoner with 3 'hops' by Peng et al. or the relation networks by Santoro et al.?
Memory networks are the most overhyped and disappointing DL "advance" I've seen yet
-
-
I studied them in depth so I could implement them from scratch and therefore teach them. Was shocked at what I found
-
What did you find?
- Još 9 drugih odgovora
Novi razgovor -
-
-
Then they took babi and replaced the names with LOTR characters and told the media the algorithm could answer questions about the book :(
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I would say GANs are overrated and overhyped not memNets
-
They may be able to help semi-supervised learning, and they may also be a useful post-processing step for "creative AI". Some potential...
Kraj razgovora
Novi razgovor -
-
-
I see it as a glorified attention mechanism. Novel for sure but not sure it's ground breaking
-
Actual attention mechanisms are much more interesting IMO.
@kchonyc's paper pre-dated memory networks, but is far (far far) more impactful - Još 1 odgovor
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.